import * as fs from 'node:fs' import * as os from 'node:os' import * as path from 'node:path' import * as url from 'node:url' import delay from 'delay' import { TimeoutError } from 'p-timeout' import type { Browser, Page, Protocol, PuppeteerLaunchOptions } from 'puppeteer' import puppeteer from 'puppeteer-extra' import RecaptchaPlugin from 'puppeteer-extra-plugin-recaptcha' import StealthPlugin from 'puppeteer-extra-plugin-stealth' import * as types from './types' puppeteer.use(StealthPlugin()) let hasRecaptchaPlugin = false let hasNopechaExtension = false const __filename = url.fileURLToPath(import.meta.url) const __dirname = url.fileURLToPath(new URL('.', import.meta.url)) /** * Represents everything that's required to pass into `ChatGPTAPI` in order * to authenticate with the unofficial ChatGPT API. */ export type OpenAIAuth = { userAgent: string clearanceToken: string sessionToken: string cookies?: Record } /** * Bypasses OpenAI's use of Cloudflare to get the cookies required to use * ChatGPT. Uses Puppeteer with a stealth plugin under the hood. * * If you pass `email` and `password`, then it will log into the account and * include a `sessionToken` in the response. * * If you don't pass `email` and `password`, then it will just return a valid * `clearanceToken`. * * This can be useful because `clearanceToken` expires after ~2 hours, whereas * `sessionToken` generally lasts much longer. We recommend renewing your * `clearanceToken` every hour or so and creating a new instance of `ChatGPTAPI` * with your updated credentials. */ export async function getOpenAIAuth({ email, password, browser, page, timeoutMs = 2 * 60 * 1000, isGoogleLogin = false, captchaToken = process.env.CAPTCHA_TOKEN, executablePath }: { email?: string password?: string browser?: Browser page?: Page timeoutMs?: number isGoogleLogin?: boolean captchaToken?: string executablePath?: string }): Promise { const origBrowser = browser const origPage = page try { if (!browser) { browser = await getBrowser({ captchaToken, executablePath }) } const userAgent = await browser.userAgent() if (!page) { page = (await browser.pages())[0] || (await browser.newPage()) page.setDefaultTimeout(timeoutMs) } await page.goto('https://chat.openai.com/auth/login', { waitUntil: 'networkidle2' }) // NOTE: this is where you may encounter a CAPTCHA if (hasRecaptchaPlugin) { await page.solveRecaptchas() } await checkForChatGPTAtCapacity(page) // once we get to this point, the Cloudflare cookies should be available // login as well (optional) if (email && password) { await waitForConditionOrAtCapacity(page, () => page.waitForSelector('#__next .btn-primary', { timeout: timeoutMs }) ) await delay(500) // click login button and wait for navigation to finish await Promise.all([ page.waitForNavigation({ waitUntil: 'networkidle2', timeout: timeoutMs }), page.click('#__next .btn-primary') ]) await checkForChatGPTAtCapacity(page) let submitP: () => Promise if (isGoogleLogin) { await page.click('button[data-provider="google"]') await page.waitForSelector('input[type="email"]') await page.type('input[type="email"]', email, { delay: 10 }) await Promise.all([ page.waitForNavigation(), await page.keyboard.press('Enter') ]) await page.waitForSelector('input[type="password"]', { visible: true }) await page.type('input[type="password"]', password, { delay: 10 }) submitP = () => page.keyboard.press('Enter') } else { await page.waitForSelector('#username') await page.type('#username', email, { delay: 20 }) await delay(100) if (hasNopechaExtension) { await waitForRecaptcha(page, { timeoutMs }) } else if (hasRecaptchaPlugin) { const res = await page.solveRecaptchas() console.log('solveRecaptchas result', res) } await delay(1200) const frame = page.mainFrame() const submit = await page.waitForSelector('button[type="submit"]', { timeout: timeoutMs }) frame.focus('button[type="submit"]') await submit.focus() await submit.click() await page.waitForSelector('#password', { timeout: timeoutMs }) await page.type('#password', password, { delay: 10 }) submitP = () => page.click('button[type="submit"]') } await Promise.all([ waitForConditionOrAtCapacity(page, () => page.waitForNavigation({ waitUntil: 'networkidle2', timeout: timeoutMs }) ), submitP() ]) } else { await delay(2000) await checkForChatGPTAtCapacity(page) } const pageCookies = await page.cookies() const cookies = pageCookies.reduce( (map, cookie) => ({ ...map, [cookie.name]: cookie }), {} ) const authInfo: OpenAIAuth = { userAgent, clearanceToken: cookies['cf_clearance']?.value, sessionToken: cookies['__Secure-next-auth.session-token']?.value, cookies } return authInfo } catch (err) { throw err } finally { if (origBrowser) { if (page && page !== origPage) { await page.close() } } else if (browser) { await browser.close() } page = null browser = null } } /** * Launches a non-puppeteer instance of Chrome. Note that in my testing, I wasn't * able to use the built-in `puppeteer` version of Chromium because Cloudflare * recognizes it and blocks access. */ export async function getBrowser( opts: PuppeteerLaunchOptions & { captchaToken?: string nopechaKey?: string } = {} ) { const { captchaToken = process.env.CAPTCHA_TOKEN, nopechaKey = process.env.NOPECHA_KEY, executablePath = defaultChromeExecutablePath(), ...launchOptions } = opts if (captchaToken && !hasRecaptchaPlugin) { hasRecaptchaPlugin = true // console.log('use captcha', captchaToken) puppeteer.use( RecaptchaPlugin({ provider: { id: '2captcha', token: captchaToken }, visualFeedback: true // colorize reCAPTCHAs (violet = detected, green = solved) }) ) } const puppeteerArgs = [ '--no-sandbox', '--disable-setuid-sandbox', '--disable-infobars', '--disable-dev-shm-usage', '--disable-blink-features=AutomationControlled', '--no-first-run', '--no-service-autorun', '--password-store=basic', '--system-developer-mode' ] if (nopechaKey) { const nopechaPath = path.join( __dirname, '..', 'third-party', 'nopecha-chrome-extension' ) puppeteerArgs.push(`--disable-extensions-except=${nopechaPath}`) puppeteerArgs.push(`--load-extension=${nopechaPath}`) hasNopechaExtension = true } const browser = await puppeteer.launch({ headless: false, // https://peter.sh/experiments/chromium-command-line-switches/ args: puppeteerArgs, ignoreDefaultArgs: [ '--disable-extensions', '--enable-automation', '--disable-component-extensions-with-background-pages' ], ignoreHTTPSErrors: true, executablePath, ...launchOptions }) // TOdO: this is a really hackity hack way of setting the API key... if (hasNopechaExtension) { const page = (await browser.pages())[0] || (await browser.newPage()) await page.goto(`https://nopecha.com/setup#${nopechaKey}`) await delay(1000) const page3 = await browser.newPage() await page.close() const extensionId = 'npgnhlnhpphdlkfdnggbdpbhoopefaai' const extensionUrl = `chrome-extension://${extensionId}/popup.html` await page3.goto(extensionUrl, { waitUntil: 'networkidle2' }) await delay(500) const editKey = await page3.waitForSelector('#edit_key .clickable') await editKey.click() const settingsInput = await page3.$('input.settings_text') await settingsInput.type(nopechaKey) await settingsInput.evaluate((el, value) => { el.value = value }, nopechaKey) await settingsInput.press('Enter') await delay(500) await editKey.click() await delay(2000) } return browser } /** * Gets the default path to chrome's executable for the current platform. */ export const defaultChromeExecutablePath = (): string => { switch (os.platform()) { case 'win32': return 'C:\\Program Files\\Google\\Chrome\\Application\\chrome.exe' case 'darwin': return '/Applications/Google Chrome.app/Contents/MacOS/Google Chrome' default: { /** * Since two (2) separate chrome releases exist on linux, we first do a * check to ensure we're executing the right one. */ const chromeExists = fs.existsSync('/usr/bin/google-chrome') return chromeExists ? '/usr/bin/google-chrome' : '/usr/bin/google-chrome-stable' } } } async function checkForChatGPTAtCapacity(page: Page) { // console.log('checkForChatGPTAtCapacity', page.url()) let res: any[] try { res = await page.$x("//div[contains(., 'ChatGPT is at capacity')]") } catch (err) { // ignore errors likely due to navigation } if (res?.length) { const error = new types.ChatGPTError('ChatGPT is at capacity') error.statusCode = 503 throw error } } async function waitForConditionOrAtCapacity( page: Page, condition: () => Promise, opts: { pollingIntervalMs?: number } = {} ) { const { pollingIntervalMs = 500 } = opts return new Promise((resolve, reject) => { let resolved = false async function waitForCapacityText() { if (resolved) { return } try { await checkForChatGPTAtCapacity(page) if (!resolved) { setTimeout(waitForCapacityText, pollingIntervalMs) } } catch (err) { if (!resolved) { resolved = true return reject(err) } } } condition() .then(() => { if (!resolved) { resolved = true resolve() } }) .catch((err) => { if (!resolved) { resolved = true reject(err) } }) setTimeout(waitForCapacityText, pollingIntervalMs) }) } async function waitForRecaptcha( page: Page, opts: { pollingIntervalMs?: number timeoutMs?: number } = {} ) { if (!hasNopechaExtension) { return } const { pollingIntervalMs = 100, timeoutMs } = opts const captcha = await page.$('textarea#g-recaptcha-response') const startTime = Date.now() if (captcha) { console.log('waiting to solve recaptcha...') do { const value = (await captcha.evaluate((el) => el.value))?.trim() if (value?.length) { // recaptcha has been solved! break } if (timeoutMs) { const now = Date.now() if (now - startTime >= timeoutMs) { throw new TimeoutError('Timed out waiting to solve Recaptcha') } } await delay(pollingIntervalMs) } while (true) } }